feat(api): add DELETE /ingest to reset the Qdrant collection

CHANGELOG.md

@@ -3,6 +3,23 @@
 All notable changes to KonjoOS are documented here.
 Format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
+## [unreleased] — Qdrant reset endpoint
+
+### Added
+- `DELETE /ingest` (`api/main.py`) — wipes the Qdrant collection `/ingest`
+  writes to and returns `{points_removed}`. Scoped narrowly: the BM25 index,
+  semantic cache, and retrieval-by-id store are untouched.
+- `QdrantStore.reset()` (`konjoai/store/qdrant.py`) — deletes and recreates
+  the collection, returning the point count it removed.
+- Motivation: `/ingest` upserts under random UUID point ids, so nothing
+  deduplicates across repeated calls — a caller re-ingesting the same corpus
+  (e.g. an external eval harness) needs a way to start from a clean,
+  single-ingest index instead of accumulating points across runs.
+
+### Tests
+- `api/test_api.py::test_reset_invokes_pipeline_reset_and_reports_points_removed`
+- `tests/unit/test_auth.py::TestQdrantStoreTenantScoping::test_reset_deletes_and_recreates_the_collection`
+
 ## [v1.7.0] — Sprint 27: Cache Warming + TTL Expiry + Query Clustering
 
 ### Added

api/main.py

@@ -10,6 +10,7 @@
     GET    /retrieval/{id}    fetch a cached retrieval result by ID
     GET    /metrics           live performance stats
     POST   /ingest            index a batch of {text, metadata} documents
+    DELETE /ingest            wipe the Qdrant collection /ingest writes to
     GET    /health            liveness check
 
 Singleflight: identical (tenant_id, question) requests collide on a single
@@ -57,6 +58,10 @@ class IngestResult(BaseModel):
     sources: int
 
 
+class ResetResult(BaseModel):
+    points_removed: int
+
+
 class QueryRequestBody(BaseModel):
     question: str = Field(..., min_length=1)
     tenant_id: str | None = Field(None, description="Tenant scope for this query.")
@@ -105,6 +110,7 @@ class Pipeline(Protocol):
     async def embed_query(self, question: str) -> np.ndarray: ...
     async def retrieve(self, question: str, q_vec: np.ndarray, top_k: int) -> list[ScoredSource]: ...
     async def index(self, request: IngestRequest) -> IngestResult: ...
+    async def reset(self) -> int: ...
     def document_count(self) -> int: ...
 
 
@@ -174,6 +180,12 @@ async def index(self, request: IngestRequest) -> IngestResult:
 
         return IngestResult(indexed=len(contents), sources=len(sources_seen))
 
+    async def reset(self) -> int:
+        from konjoai.store.qdrant import get_store
+
+        store = get_store()
+        return await asyncio.to_thread(store.reset)
+
     def document_count(self) -> int:
         try:
             from konjoai.store.qdrant import get_store
@@ -409,6 +421,22 @@ async def ingest(
             except (LookupError, ValueError):
                 pass
 
+    # ── Reset ────────────────────────────────────────────────────────────────
+
+    @app.delete("/ingest", response_model=ResetResult, tags=["ingest"])
+    async def reset_index(p: Pipeline = Depends(get_pipeline)) -> ResetResult:
+        """Wipe the Qdrant collection `/ingest` writes to.
+
+        Scoped to the vector collection only — the BM25 index, semantic
+        cache, and retrieval-by-id store are untouched. Qdrant point ids are
+        random UUIDs (see `QdrantStore.upsert`), so nothing deduplicates
+        across repeated `/ingest` calls; a caller re-ingesting the same
+        corpus (e.g. an eval harness before an authoritative run) should
+        call this first to avoid an accumulating index.
+        """
+        removed = await p.reset()
+        return ResetResult(points_removed=removed)
+
     # ── Metrics ──────────────────────────────────────────────────────────────
 
     @app.get("/metrics", response_model=MetricsResult, tags=["health"])

api/test_api.py

@@ -35,6 +35,8 @@ class _StubPipeline:
     embed_calls: int = 0
     retrieve_calls: int = 0
     index_calls: int = 0
+    reset_calls: int = 0
+    reset_points_removed: int = 0
     retrieve_delay_s: float = 0.0
     embed_dim: int = 16
 
@@ -68,6 +70,10 @@ async def index(self, request: IngestRequest) -> IngestResult:
             sources=len({d.metadata.get("source") or "default" for d in request.documents}),
         )
 
+    async def reset(self) -> int:
+        self.reset_calls += 1
+        return self.reset_points_removed
+
     def document_count(self) -> int:
         return self.document_count_value
 
@@ -198,6 +204,16 @@ async def test_ingest_rejects_empty_payload():
     assert r.status_code == 422
 
 
+@pytest.mark.asyncio
+async def test_reset_invokes_pipeline_reset_and_reports_points_removed():
+    pipeline = _StubPipeline(reset_points_removed=362)
+    async with _client(pipeline) as c:
+        r = await c.delete("/ingest")
+    assert r.status_code == 200
+    assert r.json() == {"points_removed": 362}
+    assert pipeline.reset_calls == 1
+
+
 # ── /metrics ───────────────────────────────────────────────────────────────
 
 

konjoai/store/qdrant.py

@@ -156,6 +156,26 @@ def count(self) -> int:
         """Return the number of points in the collection."""
         return self._client.count(collection_name=self._collection).count
 
+    def reset(self) -> int:
+        """Delete every point in the collection and recreate it empty.
+
+        Point ids are random UUIDs (see :meth:`upsert`), so nothing
+        deduplicates across repeated ingests — this is what gives a caller
+        (e.g. an eval harness re-ingesting the same corpus) a clean,
+        single-ingest collection instead of an accumulating one. Returns the
+        number of points removed.
+        """
+        from qdrant_client.models import Distance, VectorParams
+
+        removed = self.count()
+        self._client.delete_collection(collection_name=self._collection)
+        self._client.create_collection(
+            collection_name=self._collection,
+            vectors_config=VectorParams(size=self._dim, distance=Distance.COSINE),
+        )
+        logger.info("QdrantStore: reset collection '%s' (%d points removed)", self._collection, removed)
+        return removed
+
     def scroll_all(self, batch_size: int = 256) -> tuple[np.ndarray, list[str], list[str], list[str]]:
         """Scroll through every point in the collection.
 

tests/unit/test_auth.py

@@ -425,3 +425,24 @@ def test_search_no_filter_when_tenant_unset(self) -> None:
 
         call_kwargs = mock_client.query_points.call_args.kwargs
         assert call_kwargs.get("query_filter") is None
+
+    def test_reset_deletes_and_recreates_the_collection(self) -> None:
+        """reset() deletes+recreates the collection and returns the prior count."""
+        from konjoai.store.qdrant import QdrantStore
+
+        mock_client = MagicMock()
+        mock_client.count.return_value.count = 362
+
+        store = QdrantStore.__new__(QdrantStore)
+        store._client = mock_client
+        store._collection = "test_col"
+        store._dim = 4
+
+        removed = store.reset()
+
+        assert removed == 362
+        mock_client.delete_collection.assert_called_once_with(collection_name="test_col")
+        mock_client.create_collection.assert_called_once()
+        create_kwargs = mock_client.create_collection.call_args.kwargs
+        assert create_kwargs["collection_name"] == "test_col"
+        assert create_kwargs["vectors_config"].size == 4